Null Steering and Multi-beams Design by Complex Weight of antennas Array with the use of APSO-GA HICHEM CHAKER Department of Telecommunication University of TLEMCEN BP 2, 34 TLEMCEN, ALGERIA ALGERIA mh_chaker25@yahoo.fr Abstract: - An efficient method based on the hybrid model with breeding and subpopulations, between genetic algorithm and a modified particle swarm optimizer for the linear antenna arrays pattern synthesis with prescribed nulls in the interference direction and multi-lobe beam forming by the complex weights of each array element is presented. In general, the pattern synthesis technique that generates a desired pattern is a greatly nonlinear optimization problem. The proposed method is based on the hybrid model algorithm and the linear antenna array synthesis was modeled as a multi-objective optimization problem. Multi-objective optimization is concerned with the maximization of a vector of objectives functions in the directions of desired signal that can be subject of a number of constraints. Several numerical results with the imposed single, multiple and broad nulls sectors are provided and compared with published results to illustrate the performance of the proposed method. Key-Words: - Hybrid model, antenna array, interference suppression, multi- lobe beam forming. Introduction The antenna array pattern synthesis in order to steer nulls to the direction of interference while maintaining the main beam directed to the desired signal has received much attention [ 5]. It plays an important role in communication system, sonar, and radar applications to improve the performance (maximizing signal to interference ratio) and to cancel the jammer signal [6]. Interference suppression in antenna arrays can be achieved by steering nulls in the directions of undesired signals while keeping the main lobe in the direction of user activity by adjusting the excitation amplitude and phase. Recently, evolutionary algorithms such as particle swarm optimization (PSO) [7], simulated annealing (SA) [], and genetic algorithms (GA) [3] have been studied for array synthesis including null constrains. The evolutionary algorithms can be considered as a powerful and interesting technique for solving large kinds of electromagnetic problems. Advantages of evolutionary computation are the capability to find a global optimum, without being trapped in local optima, and the possibility to face nonlinear and discontinuous problems, with a great number of variables. On the other hand, these algorithms have strong stochastic bases, thus they require a great number of iterations to get significant results. To solve the antenna array pattern synthesis problems, among a number of optimization procedures, the artificial intelligence techniques such as genetic, simulated annealing and tabu search algorithms owing to their simplicity, flexibility and accuracy have received much attention. Genetic algorithm (GA) is a search technique based on an abstract model of Darwinian evolution. Simulated annealing (SA) technique is essentially a local search, in which a move to an inferior solution is allowed with a probability, according to some Boltzmann-type distribution, that decreases as the process progresses. Tabu search (TS) algorithm has been developed to be an effective and efficient scheme for combinatorial optimization that combines a hill-climbing search strategy based on a set of elementary moves and a heuristics to avoid to stops at sub-optimal points and the occurrence of cycles. Recently, particle swarm optimization algorithm (PSO) is proposed for solving global numerical optimization problem. The search techniques mentioned above are the probabilistic search techniques that are simple and easily be implemented without any gradient calculation. This study uses an electromagnetic optimization E-ISSN: 2224-2864 99 Volume 3, 24
technique, hybrid particle swarm optimizer with breeding and subpopulation [8]. The technique proposed in this paper is based on hybrid model algorithm to synthesis steered beams with zero in desired direction. The linear antenna array synthesis was modeled as a multi-objectives optimization problem. To verify the validity of the technique, several illustrative examples are simulated. 2 Mathematical Formulation Consider a linear array composed by 2*N equispaced isotropic antenna elements with interelement spacing λ /2. If the element excitations are conjugate-symmetrical about the center of the array, the perturbed array pattern can be written as: N 2π F( θ ) = 2 a k cos d k sin( θ ) + δ () k k = λ Where λ is the wavelength, θ denotes the angular direction, a k is the excitation amplitude, δ k is the excitation phase and d k is the x coordinate, normalised to wavelength, of the nth array element. The radiation diagram of an antenna used in our applications is determined for substrate with the permittivity equal to 3.5, thickness equal to.59 cm and operating at 5 GHz. The array factor in db is given by: P( θ ) = 2 log( F( θ ) normalised ) (2) The mathematical statement of the optimization process is: Find max f ( v) v opt (3) Where f(v) is the objective function of parameter variables v. The optimization problem can be modeled by minimization the value of difference between the perturbed and the desired patterns. Mathematically, the optimization problem can be written as: π f = Max F ( θ ) F( θ ) dθ d (4) 3 Hybrid Particle Swarm Optimizer with Breeding and Subpopulations The hybrid model incorporates one major aspect of the standard GA into the PSO, the reproduction. In the following work, we will refer to the used reproduction and recombination of genes only as breeding. Breeding is one of the core elements that make the standard GA a powerful algorithm. Hence our hypothesis was that a PSO hybrid with breeding has the potential to reach a better solution than the standard PSO. In addition to breeding we introduce a hybrid with both breeding and subpopulations. Subpopulations have previously been introduced to standard GA models mainly to prevent premature convergence to suboptimal points [9]. Our motivation for this extension was that the PSO models, including the hybrid PSO with breeding, also reach suboptimal solutions. Breeding between particles in different subpopulations was also added as an interaction mechanism between subpopulations. The traditional PSO model, described by [], consists of a number of particles moving around in the search space, each representing a possible solution to a numerical problem. Each particle has a position vector X i =(x i,,x id,,x id ), a velocity vector V i =(v i,,v id,,v id ), the position P i =(p i,,p id,,p id ) and fitness of the best point encountered by the particle, and the index (g) of the best particle in the swarm. At each iteration the velocity of each particle is updated according to their best encountered position and the best position encountered by any particle, in the following way: v id = w v id + c rand () ( pid xid ) + c2 rand () ( p x gd id ) (5) w is the inertia weight described in [] and P gd is the best position known for all particles. C and C 2 are random values different for each particle and for each dimension. If the velocity is higher than a certain limit, called V max, this limit will be used as the new velocity for this particle in this dimension, thus keeping the particles within the search space. The position of each particle is updated at each iteration. This is done by adding the velocity vector to the position vector; x id = xid + vid (6) The particles have no neighbourhood restrictions, meaning that each particle can affect all other particles. This neighbourhood is of type star (fully connected network), which has been shown to be a good neighbourhood type in [2]. Fig. shows the structure illustration of the hybrid model. Begin Initialise While (not terminate-condition) do Begin Evaluate Calculate new velocity vectors Move Breed End End Fig. The structure of the hybrid model. E-ISSN: 2224-2864 Volume 3, 24
The breeding is done by first determining which of the particles that should breed. This is done by iterating through all the particles and with probability p b (breeding probability =.6), mark a given particle for breeding. Note that the fitness is not used when selecting particles for breeding. From the pool of marked particles we now select two random particles for breeding. This is done until the pool of marked particles is empty. The parent particles are replaced by their offspring particles, thereby keeping the population size fixed. The position of the offspring is found for each dimension by arithmetic crossover on the position of the parents: Child ( i i i i 2 i x ) = p * parent ( x ) + ( p ) * parent ( x ) (7) Child 2 ( xi ) pi * parent 2( xi ) + ( pi )* parent ( xi ) = (8) Where p i is a uniformly distributed random value between and. The velocity vectors of the offspring are calculated as the sum of the velocity vectors of the parents normalized to the original length of each parent velocity vector. parent ( v) + parent ( v) Child ( v) parent 2 2 = (9) parent( v) + parent2( v) parent ( v) + parent ( v) Child ( v) parent 2 = 2 () parent( v) + parent2( v) The arithmetic crossover of positions in the search space is one of the most commonly used crossover methods with standard real valued GA, placing the offspring within the hypercube spanned by the parent particles. The main motivation behind the crossover is that offspring particles benefit from both parents. In theory this allows good examination of the search space between particles. Having two particles on different suboptimal peaks breed could result in an escape from a local optimum, and thus aid in achieving a better one. We used the same idea for the crossover of the velocity vector. Adding the velocity vectors of the parents results in the velocity vector of the offspring. Thus each parent affects the direction of each offspring velocity vector equally. In order to control that the offspring velocity was not getting too fast or too slow, the offspring velocity vector is normalized to the length of the velocity vector of one of the parent particles. The starting position of a new offspring particle is used as the initial value for this particle s best found optimum ( p i ). The motivation for introducing subpopulations is to restrict the gene flow (keeping the diversity) and thereby attempt to evade suboptimal convergence. The subpopulation hybrid PSO model is an extension of the just described breeding hybrid PSO model. In this new model the particles are divided into a number of subpopulations. The purpose of the subpopulations is that each subpopulation has its own unique best known optimum. The velocity vector of a particle is updated as before except that the best known position ( p g in the formula) now refers to the best known position within the subpopulation that the particle belongs to. In terms of the neighbourhood topology suggested by Kennedy in [], each subpopulation has its own star neighbourhood. The only interaction between subpopulations is if parents from different subpopulations breed. Breeding is now possible both within a subpopulation but also between different subpopulations. An extra parameter called probability of same subpopulation breeding p sb determines whether a given particle selected for breeding is to breed within the same subpopulation (probability p sb =.6), or with a particle from another subpopulation (probability -p sb ). Replacing each parent with an offspring particle ensures a constant subpopulation size. 4 Numerical Results To demonstrate the validity of the proposed method that synthesizes the array pattern with suppress single, multiple, and broad-band interference signal with the imposed directions and maximum tolerance of SLL using complex current excitations, several computer simulation examples using an equispaced linear array with one half wave interelement spaced 6 isotropic elements were performed. The simulation is run on hp Elitebook i5 computer with 4 GB of RAM. The algorithm of hybrid model is implemented using Matlab. The results of steering beam in the direction of the desired signal and creating single suppressed wide band interferences are presented in Figs 2, 5 and 8. E-ISSN: 2224-2864 Volume 3, 24
3 2 - -2-3 2 4 6 8 2 4 6 Fig.2 Pattern synthesis with a wide single null imposed at and steering lobe at 4. The total numbers of function evaluations is 2 iterations for this kind of excitation with 3 sub-swarms of 4 particles each one. Fig.4 The element excitation required to achieve the The array pattern synthesis is shown in Fig.5. From this figure we can see that the maximum sidelobe level is lower than -2 db. Best Subswarm 9 2 4 6 8 2 4 6 8 2 Fig.3 Convergence of the algorithm versus the number of iterations. The hybrid model APSO-AG synthesis results of amplitudes phases are traced in Fig.4 and are given in Table. Fig.5 Pattern synthesis with a wide single null imposed at 3 and steering lobe at. For the design specification of amplitude-phase synthesis APSO-AG is run for 2 generations..9 9.8.7.6.5.4 Best 8 7 6 Subswarm.3 5.2 4. 2 4 6 8 2 4 6 3 2 4 6 8 2 4 6 8 2 Fig.6 Convergence of the algorithm versus the number of iterations. E-ISSN: 2224-2864 2 Volume 3, 24
.9.8 5 Subswarm.7.6.5.4 Best -5.3 -.2. -5 3 2 4 6 8 2 4 6-2 2 4 6 8 2 4 6 8 2 Fig.9 Convergence of the algorithm versus the number ofiterations. 2.9.8.7.6 - -2.5.4-3 2 4 6 8 2 4 6 Fig.7 The element excitation required to achieve the.3.2. 2 4 6 8 2 4 6 The Fig.8 shows the normalized absolute power pattern in db, the maximum side lobe level reach - 2 db with a wide and deep single null imposed at and steering lobe at 35, we note that there is a very good agreement between desired and obtained results. The convergence of the algorithm is shown in Fig.9. 3 2 - -2-3 2 4 6 8 2 4 6 Fig. The element excitation required to achieve the The best results obtained by the hybrid model are shown in Fig., and the values are presented in Table. Fig.8 Pattern synthesis with a wide single null imposed at and steering lobe at 35. E-ISSN: 2224-2864 3 Volume 3, 24
The proposed method can create the multiple mainbeams in the directions of the different users. Figs. and 4 have shown the result of a simulation with 6 isotropic elements, cancelling interferers with -6, and two nulls at -6 and wide null at, respectively. And two steering lobes at -2 and 4. db -5 - -5-2 -25-3 -35-4 -45.9.8.7.6.5.4.3.2. 3 2 2 4 6 8 2 4 6-5 2 4 6 8 2 4 6 8 Theta Fig. The radiation pattern with a sector interference nulling around 6 and two steering lobes at 2 and 4. - -2 After 4 iterations, the fitness value reach to it maximum and the optimization process ended due to meeting the design goal. The convergence curve of fitness is presented in Fig.2. -3 2 4 6 8 2 4 6 Fig.3 The element excitation required to achieve the Best 9 8 75 7 65 Subswarm We show the comparison of the far-field patterns among the hybrid model simulation results, and the sequential quadratic programming (SQP) algorithm simulated results in [3]. An improvement of about db in the side lobe level is obtained. -5 - -5 6 5 5 2 25 3 35 4 Fig.2 Convergence of the algorithm versus the number ofiterations. db -2-25 -3-35 -4 The element excitations required to achieve the desired pattern are traced in Fig.3 and shown in Table 2. -45-5 2 4 6 8 2 4 6 8 Theta Fig.4 The radiation pattern with a wide sectors interference nulling around and two steering lobes at 2 and 4. E-ISSN: 2224-2864 4 Volume 3, 24
-5 Best 9 8 75 Subswarm db - -5-2 -25-3 7-35 -4 65-45 6 5 5 2 25 3 35 4 Fig.5 Convergence of the algorithm versus the number of iterations..9-5 2 4 6 8 2 4 6 8 Theta Fig.7 The radiation pattern with interference nulling at - 5 and tree steering lobes at 3, and 3. The corresponding number of iterations is iterations as shown in Fig.8..8.7.6.5.4.3.2 Best 9 8 75 Subswarm. 7 3 2 2 4 6 8 2 4 6 65 6 2 3 4 5 6 7 8 9 Fig.8 Convergence of the algorithm versus the number ofiterations. -.9.8.7-2 -3 2 4 6 8 2 4 6 Fig.6 The element excitation required to achieve the we introduce the cases of an array with 6 equispaced isotropic elements with λ/2 interelement spacing, which is supposed to generate three beams steered towards the three angles θ = -3, θ 2 = and θ 3 =3 with interference nulling at -5 and 5 respectively..6.5.4.3.2. 2 4 6 8 2 4 6 E-ISSN: 2224-2864 5 Volume 3, 24
3.9 2.8 -.7.6.5.4.3-2.2-3. 2 4 6 8 2 4 6 Fig.9 The element excitation required to achieve the -5 3 2 2 4 6 8 2 4 6 - db -5-2 -25-3 -35-4 - -2-3 -45-5 2 4 6 8 2 4 6 8 Theta Fig.2 The radiation pattern with interference nulling at 5 and tree steering lobes at 3, and 3. The best fitness value returned versus the number of calls to the fitness evaluator was achieved after 2. 2 4 6 8 2 4 6 Fig.22 The element excitation required to achieve the With the same array as the last section, and the same type of synthesis, we present synthesis results of multibeam array as indicated in the Fig.23, it shows the radiation pattern with four steering lobes at 3,, 4 and 38, null at -3. Subswarm Best 9 8 75 2 4 6 8 2 4 6 8 2 Fig.2 Convergence of the algorithm versus the number of iterations. Fig.23 The radiation pattern with four steering lobes at -3,,4 and 38, and null around -3. E-ISSN: 2224-2864 6 Volume 3, 24
Best 9 8 Subswarm 75 2 4 6 8 2 4 6 8 2 Fig.24 Convergence of the algorithm versus the number of iterations..9.8.7.6.5.4 optimization, where the optimization objectives are the maximum of the signal at the direction of wanted sources subject to the constraints of minimization of the signal at the direction of unwanted sources. The numerical results show that the complex excitation control using the hybrid model algorithm is efficient for prescribed single, multiple and broad nulls, and control the null depth and maximum of sidelobe level. Table. and phase distributions N Fig.2 Fig.5 Fig.8.33 97.4527.45 93.5967.886 8.986 2.5283 55.62.74 22.79.425 252.63 3.5497 292.538.2767 27.7984.5932 337.92 4.5268 66.3927.4947 98.4783.735.6828 5.55 48.7243.6246 98.3637.63 9.6975 6.735 32.4.5982 75.5256.6393 29.599 7. 77.332.794 98.944.8223 27.9489 8.7835 6.484.88 7.3346.926 26.66 9.7835-6.484.88-7.334.926-26.669. -77.33.794-98.94.8223-27.9489.735-32.4.5982-75.525.6393-29.59 2.55-48.7243.6246-98.363.63-9.697 3.5268-66.392.4947-98.478.735 -.682 4.5497-292.53.2767-27.798.5932-337.9 5.5283-55.62.74-22.79.425-252.6 6.33-97.452.45-93.596.886-8.986.3.2. 3 2-2 4 6 8 2 4 6 N Table 2. and phase distributions Fig. Fig.4 Fig.7.566 3..768 6.99.64 297.46 2.25.78.264 98.839.972 28.666 3.389 243.358.353 254.525.2533 42.448 4.4457 75.8367.5 73.7454.5272 32.88 5.434 4.696.542.62.6225 259.24 6.523 3.8.692 33.822.5755 238.75 7.637 326.4.6645 32.747.76 35.55 8.637 65.246.786 63.344.823 26.549 9.637-65.24.786-63.34.823-26.54.637-326.4.6645-32.74.76-35.5.523-3.88.692-33.82.5755-238.7 2.434-4.69.542 -.6.6225-259.24 3.4457-75.836.5-73.745.5272-32.88 4.389-243.35.353-254.52.2533-42.44 5.25 -.7.264-98.83.972-28.66 6.566-3..768-6.99.64-297.4-2 Table 3. and phase distributions -3 2 4 6 8 2 4 6 Fig.25 The element excitation required to achieve the The best results obtained by the hybrid model are listen in table 3. 5 Conclusion A method for antenna array pattern synthesis based on the hybrid model algorithm has been presented. The problem can be modeled as a multicriteria N Fig.2 Fig.23.592 92.8.244 248.59 2.2465 2.762.2678 52.64 3.3496 252.9.5634 2.22 4.456 267.748.69 236. 5.5649 47.559.2592 34.439 6.6924 22.2.6237.34 7.882 24.679.56 4.367 8.7826 235.972.84 8.7 9.7826-235.972.84-8.7.882-24.679.56-4.367.6924-22.2.6237 -.34 2.5649-47.559.2592-34.439 3.456-267.748.69-236. 4.3496-252.9.5634-2.22 5.2465-2.762.2678-52.64 6.592-92.8.244-248.59 E-ISSN: 2224-2864 7 Volume 3, 24
References: [] V. Murino, A. Trucco and C.S. Regazzoni, Synthesis of Equally Spaced Arrays by Simulated Annealing, IEEE Trans Signal Process 44,pp.9 22, 996. [2] R.L. Haupt, Phase-only Adaptive Nulling with a Genetic Algorithm, IEEE Trans Antennas Propag, 45, pp.9 5. 997. [3] W.P. Liao and F.L.Chu, Array Pattern Nulling by Phase and Position Perturbations with the use of the Genetic Algorithm, Microwave Opt Technol Lett, 5, pp. 25 256, 997. [4] R. Vescovo, Null Synthesis by Phase Control for Antenna Array, Electronic Letters, 36,pp.98 99. 2. [5] K. Gunney and A. Akdagli, Null Steering of Linear Antenna Arrays using a Modified Tabu Search Algorithm, Prog Electromagn Res, 33, pp. 67 82, 2. [6] N. Karaboga, K. Gunney and A. Akdagli, Null Steering of Linear Antenna Arrays with use of Modified Touring ant Colony Optimization Algorithm, Wiley Periodicals; Int J RF Microwave Computer-Aided Eng, 2, pp.375 383. 22. [7] M.M. Khodier and C.G. Christodoulou, Linear Array Geometry Synthesis with Minimum Sidelobe Level and Null Control using Particle Swarm Optimization, IEEE Trans Antennas Propag. 53, pp.2674 2679, 25. [8] M. Lovbjerg, T.K. Rasmussen and T. Krink, Hybrid Particle Swarm Optimizer with Breeding and Subpopulations, In Proceedings of the Genetic and Evolutionary Computation Conference, San Francisco. 2. [9] W.M. Spears, Simple Subpopulation Schemes, proceeding of the evolutionary programming conference, pp.296-3, 994. [] J. Kennedy and R. C. Eberhart, Particle Swarm Optimization, IEEE proceeding of international conference on neural networks, vol.4, pp. 942-948, 9. [] Y. Shi and R. C. Eberhart, A Modified Particle Swarm Optimiser, IEEE international conference on evolutionary computation, Alaska, May 998. [2] Y. Shi and R. C. Eberhart, Parameter selection in particle swarm optimization, lecture notes in computer science (Evolutionary programming VII, pp. 59-6. Springer. 998. [3] M. Mouhamadou and P. Vaudon, Smart antenna array patterns synthesis: Null steering and multi-user beamforming by phase control, Progress In Electromagnetics Research, PIER, 6, pp. 6, 26. was born in Ghazaouet ALGERIA in 98. He graduated in electronics from the University of Tlemcen in 24, where he also received the master and Ph.D degrees in electronics (signals and systems) in 28 and 22 respectively, from 29 to 2, he worked in the department of electronic at the university of Tlemcen as an associate researcher. Since April 2, he has been working with vimpelcom as Siemens radio frequency planning and optimization senior engineer at ALGERIA. His research interests include micro strip antenna arrays simulation and modelling. E-ISSN: 2224-2864 8 Volume 3, 24